Comparison of STOI-type Intermediate Feature for Listening Difficulty Rating Prediction Model

Automatic quality evaluation of outdoor public-address system is required for a disaster situation. We had earlier developed a system to predict the listening difficulty rating (LDR) of a PA system. However, the prediction error was large, and the performance was not satisfactory. Here, we use STOI-type indices as intermediate features for improvement of the LDR prediction system. The results show that ESTOI and SIMI should be chosen optimally for each SNR condition.